Method for discarding corrupted data packets in a reliable transport fabric

Abstract

A method for discarding perpetually-rejected packets in a fabric-based interconnect having a reliable physical layer is disclosed. A transmitting component keeps a count of the number of negative acknowledgements (NAKs) it receives from the receiving component for packets the transmitting component sends. If the transmitting component receives a number of consecutive NAKs for the same packet that exceeds some pre-determined threshold, the packet is not resent, but is, instead, treated as having been acknowledged, and subsequent packets are allowed to be transmitted. Higher-level processes are then notified of the problem so as to allow the error to be dealt with at a higher level, but without obstructing the flow of packets on the physical layer.

Description

TECHNICAL FIELD[0001]The present invention relates generally to fabric-based interconnects in data processing systems. Specifically, the present invention is directed to a method of fault recovery for use in a fabric-based interconnect having a reliable physical layer.BACKGROUND ART[0002]Electronic systems typically rely on buses to transfer data between components. A bus is a signal route to which system components are connected in parallel so that signals can be passed between them. Although buses are relatively convenient from an implementation standpoint, the bus paradigm has a number of drawbacks. First, because buses connect multiple components in parallel, much time must be spent arbitrating between different components wishing to access the bus. Second, traditional bus systems typically do not allow a user to add or remove a component to / from the system while the system is operating, due to the fact that all of the components on the bus are connected electrically to each oth...

AI Technical Summary

Benefits of technology

[0008]A preferred embodiment of the present invention provides a method for discarding perpetually-rejected packets in a fabric-based interconnect having a reliable physical layer. A transmitting component keeps a count of the number of negative acknowledgements (NAKs) it receives from the receiving component for packets the transmitting component sends. If the transmitting component receives a numbe

Problems solved by technology

First, because buses connect multiple components in parallel, much time must be spent arbitrating between different components wishing to access the bus.

Second, traditional bus systems typically do not allow a user to add or remove a component to / from the system while the system is operating, due to the fact that all of the components on the bus are connected electrically to each other in parallel.

Advantages of this approach are that it obviates the need for costly (in terms of performance) bus arbitration protocol and that it makes it possible to “hot-swap” components (i.e., connect or disconnect components while the system is operating).

This can cause a problem, because it sometimes happens that a packet is repeatedly rejected, due to some corruption of the packet itself or unexpected change of operating conditions (e.g., if a RAPIDIO™ device starts rejecting specific classes of packets based on a configuration bit).

This process can have a devastating effect on system performance, due to the fact that entire classes of packets are stalled within the system until the expiration of some timeout.

Images